[sorry for a reply from -digest]
feel free to critisize
Dominique Muller wrote:
> A propos de la conf�rence d'hier soir, est-ce qu'il est pr�vu de faire un
> r�sum� pour ceux comme moi, int�ress�, mais n'ayant pu venir ?
Evolution r�cente du noyau Linux
Prof. Daniel P. Bovet
Gen�ve, Uni Mail, Amphi R060, mercredi 9 janvier 2002 � 20h30
talk notes
[EMAIL PROTECTED]
11 jan 2002
Efficient OS
============
* same hardware, different OS
* applications (even compiler) don't make a better OS
* better kernel => better performance
Two faces of a kernel
---------------------
* interface: interaction with hardware from user space
* performance: optimal use of available hardware (hardware evolves rapidly)
How to obtain good performance ?
--------------------------------
* there is no magic
* many minor improvements make the difference
Linux evolution
===============
* new hardware => need for new software solutions (!<=)
* a lot of redesign
* experimental and stable kernels
2.2 -> 2.4 (3 good points)
--------------------------
1. TLB (translation lookaside buffer)
TLB is used to reduce the number of calls to obtain mem. addr in RAM
(x86: CR3 -> page directory -> page table -> page frame -> addr)
improved TLB management during task (process) switching
new flag for kernel threads
2. task scheduler
when CPU "has nothing to do", it "runs" an idle process
each process has a dynamic priority
schedule() selects process with highest priority for a given CPU
linux scheduling is not preemptive (kernel threads can't be interrupted)
when schedule() should be called?
process blocks (finishes?)
process exec time has expired (time sharing) ~ 10-100 ms
process with higher priority wakes up
sched_yield() or sched_setscheduler() been called
(more difficulties with multi-CPU machines, IPI: InterProcessorInterrupt)
3. memory (page frame) reclaiming (linux specific)
several software caches
caches size is not fixed (can grow until no more free memory)
frame reclaiming = antidote from system freezing (lack of free memory)
pages have 3 types of priorities (low -> high)
2.2 kernel swap out
penalize a process which uses the largest nb. of pages
(not good if memory is used effectively)
2.4 improvement
2 LRU(?) lists: active list (pages used frequently) and inactive list
lists are updated dynamically (how to balance the lists?)
<a href="http://bravo.ce.uniroma2.it/kernelhacking/en/index.html";>
linux kernel hacking</a>
--
Ivan Zimine | [EMAIL PROTECTED]
Dpt. of Radiology | (+41 22) 372 70 70
Geneva University Hospitals |
--
http://www-internal.alphanet.ch/linux-leman/ avant de poser
une question. Ouais, pour se d�sabonner aussi.
